Abstract: A process for the automatic detection of defects in a complexe part comprises segmenting a radioscopic image of the part into two classes of zone in such a way as to obtain a binary images referred to as a mask, in which the assessable zones and the non-assessable zones respectively assume the values 1 and 0, and applying to the radioscopic image conditional morphological transformations formulated from a conditional formative element obtained by intersection between a conventional formative element and the mask. The process is particularly applicable to the radioscopic inspection and evaluation of blade castings.

Abstract: A process for automatic detection of the assessable areas within an image of a mechanical component comprises a first phase during which different areas of at least one reference image are detected by means of manual segmentation, a second phase during which marking parameters for the areas of the reference images are defined and fixed, these marking parameters being simplifying transformations and thresholding intervals for the reference images making it possible to obtain optimal quality of separation of the areas, and a fully automatic third phase during which the area-marking parameters defined and fixed during the first two phases are used to mark the assessable areas of new images, and the exact outline of said areas is determined using the watershed line method.